1 Letter

I am writing to share my thoughts regarding the data analysis employed by Digby et al. (2023). Using a between-subjects study design, Digby et al. (2023) investigated the response of peak oxygen uptake to cycling and running sprint interval training in 2 well-balanced groups of 9 young adults each, with sessions performed three times per week for three consecutive weeks. Regardless of the exercise mode, training sessions involved 30-s all-out sprints interspersed with 4-min resting periods, as well as a 5-min warm-up and cool-down at participants’ preferred intensity. While 4 sprints per session were performed in the first week of training, a fifth and a sixth sprint were added to the sessions of weeks 2 and 3, respectively. A relevant aspect of the study was the measurement of adaptive responses using both a cycle ergometer and a treadmill, which allowed Digby et al. (2023) to examine crossover effects when tests are conducted with exercise modes that do not correspond with those of training. This study showed that cycling and running sprint interval training are equally effective in improving peak oxygen uptake, with mean increases of 5.2% and 5.6%, respectively. These improvements were nevertheless abolished when testing modes did not coincide with training modes, providing a compelling demonstration of the specificity principle (Reilly et al., 2009).

Alongside the usual cohort-based analysis, Digby et al. (2023) determined individual response rates (%) for peak oxygen uptake within each group. Essentially, participants were categorised as responders or non-responders depending on whether their change in peak oxygen uptake exceeded a threshold, calculated as \(2 \times SD_{diff}/\sqrt{2}\). Here, \(SD_{diff}\) represents the standard deviation of the differences between the first and second measurements across participants in a reliability study. According to Hopkins (2000), cited by Digby et al. (2023), this procedure could be used for the monitoring of an individual (see page 7). However, Hopkins (2000) also dedicated a section to the estimation of the magnitude of individual differences (see page 9), where he presents an equation to gauge inter-individual variability in the experimental group(s) based on the variability associated with the control group or a reliability study. This approach operates under the premise that systematic effects can be attributed to the intervention only when outcomes in the absence of such intervention are accounted for—although relying on a shorter reliability study may inflate inter-individual variability estimates, which Hopkins (2000) failed to acknowledge. Despite Hopkins (2000) presenting a somewhat contradictory and incomplete perspective, Digby et al. (2023) unequivocally asserted that a control group was not required for their study. This wrong assumption, as explained below, indicates a potential oversight or deliberate disregard for the recommendations of many other academics advocating for specific experimental designs and statistical analyses to investigate inter-individual variability (Atkinson et al., 2019; Atkinson & Batterham, 2015; Chrzanowski-Smith et al., 2020; Egger et al., 2016; Hecksteden et al., 2015, 2018; Hrubeniuk et al., 2022; Ross et al., 2019; Senn, 2016; Voisin et al., 2019; Williamson et al., 2017).

Reliability research in exercise sciences has primarily concentrated on measurement errors and day-to-day biological fluctuations in response to a period without intervention (Hopkins, 2000). However, the sources of variability associated with an intervention are more difficult to quantify and control for (Atkinson et al., 2019; Atkinson & Batterham, 2015; Chrzanowski-Smith et al., 2020; Del Giudice et al., 2020; Egger et al., 2016; Hecksteden et al., 2015, 2018; Hrubeniuk et al., 2022; Jamnick et al., 2020; Mann et al., 2013; Meyler et al., 2021; Ross et al., 2019; Senn, 2016; Voisin et al., 2019; Williamson et al., 2017). Indeed, a close examination of some old publications might suggest that within-individual adaptive responses are hardly replicable when endurance training programmes are repeated on distinct occasions (Cureton & Phillips, 1964; Örlander et al., 1977; Pedersen & Jørgensen, 1978; Simoneau et al., 1987); a hypothesis corroborated by more recent work (Del Giudice et al., 2020). While these training studies may have failed to standardise the training load between and within individuals, given the limitations of relative intensity prescription methods (Egger et al., 2016; Jamnick et al., 2020; Mann et al., 2013; Meyler et al., 2021), their results reflect the difficulty in accurately estimating individual adaptive responses. As a minimum, it requires that training responses are measured at multiple points along the course of an intervention for modelling purposes, or the inclusion of a control group/period for comparative analysis based on standard deviations (Atkinson et al., 2019; Atkinson & Batterham, 2015; Chrzanowski-Smith et al., 2020; Hecksteden et al., 2015, 2018; Hrubeniuk et al., 2022; Ross et al., 2019; Senn, 2016; Voisin et al., 2019; Williamson et al., 2017). These methodological approaches account, at least partially, for the random measurement error and long-term biological fluctuations typical of training studies, in stark contrast with the simplistic post-hoc analysis conducted by Digby et al. (2023).

While I appreciate the hard work and dedication that Digby et al. (2023) have put into their study, effectively demonstrating the specificity principle in the context of sprint interval training, I believe that the (non-)responder categorisation adds nothing to their main findings. Objectively, my view is that the raw analysis of individual responses is fundamentally flawed. It overlooks the potential impact of compounded measurement errors and biological fluctuations associated with testing, both in the presence and absence of an intervention. It also neglects the possibility that within-individual adaptive responses are inherently variable from one training intervention to another. Put simply, there is no guarantee that the observed changes in peak oxygen uptake in the study of Digby et al. (2023) represent the true responses for each individual. In reality, the number of responders and non-responders within a sample is strongly influenced by the magnitude of the mean adaptive response, regardless of the response threshold chosen, as demonstrated nicely by simulations (https://ahbossi.shinyapps.io/Bossi_Simulator). Therefore, authors should avoid using this approach in an attempt to enhance the impact of their studies, as it could potentially detract readers from appreciating their primary findings. When the investigation of inter-individual variability in adaptive responses is the main research focus, an a priori decision should be made to use appropriate experimental designs and statistical analyses, for which a vast body of literature is available (Atkinson et al., 2019; Atkinson & Batterham, 2015; Chrzanowski-Smith et al., 2020; Hecksteden et al., 2015, 2018; Hrubeniuk et al., 2022; Ross et al., 2019; Senn, 2016; Voisin et al., 2019; Williamson et al., 2017).

2 Concluding Remarks and Future Directions

While this letter is deliberately succinct, readers are likely to benefit further from the literature previously referenced. My primary aim was to summarise the pitfalls associated with the raw analysis of individual responses, rather than providing a comprehensive review. Accompanying this letter is an interactive beta-version simulator, which I propose should undergo further optimisation to account for all sources of within-individual variability affecting the adaptive responses to training measured at the participant level. Future versions should also allow users to select from a variety of experimental designs to observe the impact on the (non-)responder categorisation, whilst presenting methodologically sound alternatives for the investigation of inter-individual variability. The source code for the simulator is publicly available (https://doi.org/10.5281/zenodo.8387674), and contributions to its continued development are highly encouraged. This is the first step in what must be a collective educational effort to enhance the rigour and validity of individual response research in exercise sciences.

3 Additional Information

4 References